src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/legacyheaders/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: CubicSplineTable
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_VF_sparc64_hpc_ace_double
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB;
  76     int              j_coord_offsetA,j_coord_offsetB;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  81     int              vdwioffset0;
  82     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  83     int              vdwjidx0A,vdwjidx0B;
  84     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  85     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  93     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  94     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  95     real             *vftab;
  96     _fjsp_v2r8       itab_tmp;
  97     _fjsp_v2r8       dummy_mask,cutoff_mask;
  98     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  99     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 100     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 101
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_elec->data;
 120     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _fjsp_setzero_v2r8();
 148         fiy0             = _fjsp_setzero_v2r8();
 149         fiz0             = _fjsp_setzero_v2r8();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _fjsp_setzero_v2r8();
 157         vvdwsum          = _fjsp_setzero_v2r8();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 175             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 176             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 182
 183             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 198             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 199                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 200
 201             /* Calculate table index by multiplying r with table scale and truncate to integer */
 202             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 203             itab_tmp         = _fjsp_dtox_v2r8(rt);
 204             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 205             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 206             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 207
 208             vfconv.i[0]     *= 4;
 209             vfconv.i[1]     *= 4;
 210
 211             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 212             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 213             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 214             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 215             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 216             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 217             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 218             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 219             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 220             velec            = _fjsp_mul_v2r8(qq00,VV);
 221             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 222             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 223
 224             /* LENNARD-JONES DISPERSION/REPULSION */
 225
 226             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 227             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 228             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 229             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 230             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 231
 232             /* Update potential sum for this i atom from the interaction with this j atom. */
 233             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 234             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 235
 236             fscal            = _fjsp_add_v2r8(felec,fvdw);
 237
 238             /* Update vectorial force */
 239             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 240             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 241             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 242
 243             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 244
 245             /* Inner loop uses 59 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             jnrA             = jjnr[jidx];
 252             j_coord_offsetA  = DIM*jnrA;
 253
 254             /* load j atom coordinates */
 255             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 260             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 261             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 267
 268             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 272             vdwjidx0A        = 2*vdwtype[jnrA+0];
 273
 274             /**************************
 275              * CALCULATE INTERACTIONS *
 276              **************************/
 277
 278             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 282             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 283                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 284
 285             /* Calculate table index by multiplying r with table scale and truncate to integer */
 286             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 287             itab_tmp         = _fjsp_dtox_v2r8(rt);
 288             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 289             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 290             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 291
 292             vfconv.i[0]     *= 4;
 293             vfconv.i[1]     *= 4;
 294
 295             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 296             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 297             F                = _fjsp_setzero_v2r8();
 298             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 299             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 300             H                = _fjsp_setzero_v2r8();
 301             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 302             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 303             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 304             velec            = _fjsp_mul_v2r8(qq00,VV);
 305             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 306             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 307
 308             /* LENNARD-JONES DISPERSION/REPULSION */
 309
 310             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 311             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 312             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 313             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 314             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 318             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 319             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 320             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 321
 322             fscal            = _fjsp_add_v2r8(felec,fvdw);
 323
 324             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 325
 326             /* Update vectorial force */
 327             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 328             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 329             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 330
 331             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 332
 333             /* Inner loop uses 59 flops */
 334         }
 335
 336         /* End of innermost loop */
 337
 338         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 339                                               f+i_coord_offset,fshift+i_shift_offset);
 340
 341         ggid                        = gid[iidx];
 342         /* Update potential energies */
 343         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 344         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 345
 346         /* Increment number of inner iterations */
 347         inneriter                  += j_index_end - j_index_start;
 348
 349         /* Outer loop uses 9 flops */
 350     }
 351
 352     /* Increment number of outer iterations */
 353     outeriter        += nri;
 354
 355     /* Update outer/inner flops */
 356
 357     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
 358 }
 359 /*
 360  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 361  * Electrostatics interaction: CubicSplineTable
 362  * VdW interaction:            LennardJones
 363  * Geometry:                   Particle-Particle
 364  * Calculate force/pot:        Force
 365  */
 366 void
 367 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 368                     (t_nblist                    * gmx_restrict       nlist,
 369                      rvec                        * gmx_restrict          xx,
 370                      rvec                        * gmx_restrict          ff,
 371                      t_forcerec                  * gmx_restrict          fr,
 372                      t_mdatoms                   * gmx_restrict     mdatoms,
 373                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 374                      t_nrnb                      * gmx_restrict        nrnb)
 375 {
 376     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 377      * just 0 for non-waters.
 378      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 379      * jnr indices corresponding to data put in the four positions in the SIMD register.
 380      */
 381     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 382     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 383     int              jnrA,jnrB;
 384     int              j_coord_offsetA,j_coord_offsetB;
 385     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 386     real             rcutoff_scalar;
 387     real             *shiftvec,*fshift,*x,*f;
 388     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 389     int              vdwioffset0;
 390     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 391     int              vdwjidx0A,vdwjidx0B;
 392     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 393     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 394     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 395     real             *charge;
 396     int              nvdwtype;
 397     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 398     int              *vdwtype;
 399     real             *vdwparam;
 400     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 401     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 402     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 403     real             *vftab;
 404     _fjsp_v2r8       itab_tmp;
 405     _fjsp_v2r8       dummy_mask,cutoff_mask;
 406     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 407     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 408     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 409
 410     x                = xx[0];
 411     f                = ff[0];
 412
 413     nri              = nlist->nri;
 414     iinr             = nlist->iinr;
 415     jindex           = nlist->jindex;
 416     jjnr             = nlist->jjnr;
 417     shiftidx         = nlist->shift;
 418     gid              = nlist->gid;
 419     shiftvec         = fr->shift_vec[0];
 420     fshift           = fr->fshift[0];
 421     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 422     charge           = mdatoms->chargeA;
 423     nvdwtype         = fr->ntype;
 424     vdwparam         = fr->nbfp;
 425     vdwtype          = mdatoms->typeA;
 426
 427     vftab            = kernel_data->table_elec->data;
 428     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 429
 430     /* Avoid stupid compiler warnings */
 431     jnrA = jnrB = 0;
 432     j_coord_offsetA = 0;
 433     j_coord_offsetB = 0;
 434
 435     outeriter        = 0;
 436     inneriter        = 0;
 437
 438     /* Start outer loop over neighborlists */
 439     for(iidx=0; iidx<nri; iidx++)
 440     {
 441         /* Load shift vector for this list */
 442         i_shift_offset   = DIM*shiftidx[iidx];
 443
 444         /* Load limits for loop over neighbors */
 445         j_index_start    = jindex[iidx];
 446         j_index_end      = jindex[iidx+1];
 447
 448         /* Get outer coordinate index */
 449         inr              = iinr[iidx];
 450         i_coord_offset   = DIM*inr;
 451
 452         /* Load i particle coords and add shift vector */
 453         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 454
 455         fix0             = _fjsp_setzero_v2r8();
 456         fiy0             = _fjsp_setzero_v2r8();
 457         fiz0             = _fjsp_setzero_v2r8();
 458
 459         /* Load parameters for i particles */
 460         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 461         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 462
 463         /* Start inner kernel loop */
 464         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrA             = jjnr[jidx];
 469             jnrB             = jjnr[jidx+1];
 470             j_coord_offsetA  = DIM*jnrA;
 471             j_coord_offsetB  = DIM*jnrB;
 472
 473             /* load j atom coordinates */
 474             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 475                                               &jx0,&jy0,&jz0);
 476
 477             /* Calculate displacement vector */
 478             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 479             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 480             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 481
 482             /* Calculate squared distance and things based on it */
 483             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 484
 485             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 486
 487             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 502             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 503                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 504
 505             /* Calculate table index by multiplying r with table scale and truncate to integer */
 506             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 507             itab_tmp         = _fjsp_dtox_v2r8(rt);
 508             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 509             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 510             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 511
 512             vfconv.i[0]     *= 4;
 513             vfconv.i[1]     *= 4;
 514
 515             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 516             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 517             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 518             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 519             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 520             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 521             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 522             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 523             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 524             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 525
 526             /* LENNARD-JONES DISPERSION/REPULSION */
 527
 528             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 529             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 530
 531             fscal            = _fjsp_add_v2r8(felec,fvdw);
 532
 533             /* Update vectorial force */
 534             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 535             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 536             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 537
 538             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 539
 540             /* Inner loop uses 50 flops */
 541         }
 542
 543         if(jidx<j_index_end)
 544         {
 545
 546             jnrA             = jjnr[jidx];
 547             j_coord_offsetA  = DIM*jnrA;
 548
 549             /* load j atom coordinates */
 550             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 551                                               &jx0,&jy0,&jz0);
 552
 553             /* Calculate displacement vector */
 554             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 555             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 556             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 557
 558             /* Calculate squared distance and things based on it */
 559             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 560
 561             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 562
 563             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 564
 565             /* Load parameters for j particles */
 566             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 567             vdwjidx0A        = 2*vdwtype[jnrA+0];
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 577             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 578                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 579
 580             /* Calculate table index by multiplying r with table scale and truncate to integer */
 581             rt               = _fjsp_mul_v2r8(r00,vftabscale);
 582             itab_tmp         = _fjsp_dtox_v2r8(rt);
 583             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 584             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 585             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 586
 587             vfconv.i[0]     *= 4;
 588             vfconv.i[1]     *= 4;
 589
 590             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 591             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 592             F                = _fjsp_setzero_v2r8();
 593             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 594             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 595             H                = _fjsp_setzero_v2r8();
 596             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 597             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 598             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 599             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,FF),_fjsp_mul_v2r8(vftabscale,rinv00)));
 600
 601             /* LENNARD-JONES DISPERSION/REPULSION */
 602
 603             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 604             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 605
 606             fscal            = _fjsp_add_v2r8(felec,fvdw);
 607
 608             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 609
 610             /* Update vectorial force */
 611             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 612             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 613             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 614
 615             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 616
 617             /* Inner loop uses 50 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 623                                               f+i_coord_offset,fshift+i_shift_offset);
 624
 625         /* Increment number of inner iterations */
 626         inneriter                  += j_index_end - j_index_start;
 627
 628         /* Outer loop uses 7 flops */
 629     }
 630
 631     /* Increment number of outer iterations */
 632     outeriter        += nri;
 633
 634     /* Update outer/inner flops */
 635
 636     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);
 637 }